16
INTRODUCTION TO FIXED POINT AND FLOATING POINT PROCESSORS Presented by: Sarbjeet Singh NITTTR- Chandigarh

fixed point and floating point processors

Embed Size (px)

Citation preview

Page 1: fixed point and floating point processors

INTRODUCTION TOFIXED POINT AND FLOATING POINT

PROCESSORS

Presented by:Sarbjeet SinghNITTTR-Chandigarh

Page 2: fixed point and floating point processors

DSP PROCESSORS DATA FORMAT

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS22

Fixed point represent each no by with the minimum of 16 bits.

Floating point take 32 bits to store each value

Page 3: fixed point and floating point processors

FIXED POINT FORMAT

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS33

There are four common ways that these 2^16 = There are four common ways that these 2^16 = 65,536 possible bit patterns can represent a number65,536 possible bit patterns can represent a number

Page 4: fixed point and floating point processors

FIXED POINT DATA FORMAT UNSIGNED INTEGER o The stored no can take one of any integer

value from 0 to 65536 SIGNED INTEGER o Use two’s compliment to make the rangeo Include negative numbers that is from -

32768 to 32767 UNSIGNED FRACTION NOTATIONo The 65536 levels are spread uniformly

between 0 and 1 SIGNED FRACTION FORMAT

ALLOWSo Negative no, equally spaced between -1 to 1 04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS44

Page 5: fixed point and floating point processors

FLOATING POINT FORMAT use a minimum of 32 bits to store each value2^32 =4,294,967,296This results in many more bit patterns than

for fixed pointA key feature of floating point notation is that the represented numbers are not uniformly spaced. All floating point DSPs can also handle fixed

point numbers e.g.- SHARC processor

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS55

Page 6: fixed point and floating point processors

FIXED POINT PROCESSORS Fixed point processor are……….

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS66

Cheaper and less power Cheaper and less power consumingconsumingUse shorter data word width that will Use shorter data word width that will provide adequate accuracyprovide adequate accuracyMost of fixed point DSP processor use 16 Most of fixed point DSP processor use 16 bit data wordsbit data words•SSufficient for many applicationsufficient for many applicationsMost DSP includes one or more Most DSP includes one or more accumulator registersaccumulator registers•Wider than other registersWider than other registers•Provide extra guard bits to avoid Provide extra guard bits to avoid overflowoverflow

Page 7: fixed point and floating point processors

FIXED VERSUS FLOATING POINT DSP

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS77

While floating While floating point devices point devices have have •Better precision, Better precision, •higher dynamic higher dynamic RangeRange•shorter shorter development development cycle.cycle.

Fixed point DSPs Fixed point DSPs are generally are generally

cheapercheaper

Page 8: fixed point and floating point processors

FIXED POINT VS FLOATING POINT

Fixed Point/Floating PointFIXED POINT PROCESSOR are :

cheapersmallerless power consumingHarder to program

Watch for errors: truncation, overflow, rounding

Limited dynamic rangeUsed in 95% of consumer products

FLOATING POINT PROCESSORShave larger accuracyare much easier to programcan access larger memory

It is harder to create an efficient program in CC on a fixed point processors than on floating point processors 04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS88

Page 9: fixed point and floating point processors

PERFORMANCE

Provide

HIGH SIGNAL TO NOISE RATIO

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS99

WHAT 32 BIT FLOATING POINT WHAT 32 BIT FLOATING POINT SYSTEM CAN DO THAT 16 BIT SYSTEM CAN DO THAT 16 BIT FIXED POINT CAN’T FIXED POINT CAN’T ??????????????????????????????????????

Page 10: fixed point and floating point processors

the signal-to-noise ratio for storing a floating point number is

about 30 million to one while for a fixed point number it is only about ten-thousand to one

floating point has roughly 3,000 times less quantization noise than fixed point

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1010

Page 11: fixed point and floating point processors

Fixed versus floating point instructions These are the multiplication instructions

used inthe SHARC DSPs. While only a single command is

needed for floating point, many options are needed for fixed point.

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1111

Page 12: fixed point and floating point processors

FIXED POINT VS FLOATING POINTFloating Point Fixed Point

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1212

ApplicationsApplications

•Modems

•Digital Subscriber Line (DSL)

•Wireless Base stations

•Central Office Switches

•Private Branch Exchange (PBX)

•Digital Imaging

•3D Graphics

•Speech Recognition

•Voice over IP

ApplicationsApplications

•Portable Products

•2G, 2.5G and 3G Cell Phones

•Digital Audio Players

•Digital Still Cameras

•Electronic Books

•Voice Recognition

•GPS Receivers

•Headsets

•Biometrics

•Fingerprint Recognition

Page 13: fixed point and floating point processors

DIGITAL SIGNAL PROCESSORS COMPANIES

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1313

ANALOG DEVICESANALOG DEVICES•ADSP-21xx-16 bit, ADSP-21xx-16 bit, fixed pointfixed point•ADSP-21xxx-32 bit, ADSP-21xxx-32 bit, floating point and fixed floating point and fixed pointpoint

LUCENT LUCENT TECHNOLOGIESTECHNOLOGIES•DSP 16xxx-16 bit, DSP 16xxx-16 bit, fixed pointfixed point•DSP 32xx-32 bit, DSP 32xx-32 bit, floating pointfloating point

MOTOROLAMOTOROLA•DSP 561xx,16 bit, DSP 561xx,16 bit, fixed pointfixed point•DSP 560xx,24 bit, DSP 560xx,24 bit, fixed pointfixed point•DSP 96002,32 bit, DSP 96002,32 bit, floating pointfloating point

TEXAS TEXAS INSTRUMENTSINSTRUMENTS•TMS 320cxx,16 bit TMS 320cxx,16 bit fixed pointfixed point•TMS 320cxxx,32 bit, TMS 320cxxx,32 bit, floating pointfloating point

Page 14: fixed point and floating point processors

CONCLUSIONfloating point and fixed point usually use 32

bits and 16 bits, respectively, but not always.

Fixed point processor are :Cheaper , smaller , less power consuming ,

Harder to programWatch for errors: truncation, overflow,

rounding Limited dynamic range

Floating point processors have larger accuracy , are much easier to program , can access larger memory

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1414

Page 15: fixed point and floating point processors

REFERENCEI feachor & Jervis - Digital Signal ProcessingThe Scientist and Engineer's Guide to

Digital Signal Processinghttp://www.sundance.com/index.htmhttp://www.bdti.com/

04/12/2304/12/23FIXED POINT AND FLOATING POINT PROCESSORSFIXED POINT AND FLOATING POINT PROCESSORS1515

Page 16: fixed point and floating point processors

04/12/2304/12/231616